Related Experiment Video
Updated: Aug 30, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.3K
How much BiGAN and CycleGAN-learned hidden features are effective for COVID-19 detection from CT images? A
Sima Sarv Ahrabi1, Alireza Momenzadeh1, Enzo Baccarelli1
1Department of Information Engineering, Electronics and Telecommunications, Sapienza University or Rome, Via Eudossiana, 18, 00184 Roma, Italy.
The Journal of Supercomputing
|August 31, 2022
Summary
New machine learning models, Bidirectional generative adversarial networks (BiGANs) and CycleGANs, can detect COVID-19 from CT scans by analyzing hidden features. These models significantly outperform existing methods in classification accuracy for lung disease detection.
Area of Science:
- Artificial Intelligence
- Medical Imaging Analysis
- Machine Learning
Background:
- Bidirectional generative adversarial networks (BiGANs) and CycleGANs are primarily used as generative models.
- Their encoding modules offer potential for extracting hidden features from input data.
- Exploiting these features for classification tasks remains an underexplored area.
Purpose of the Study:
- To develop and test a novel inference engine utilizing BiGAN and CycleGAN features for COVID-19 detection in CT scans.
- To compare the performance of BiGAN and CycleGAN models against state-of-the-art Convolutional Autoencoder (CAE) methods.
- To evaluate classification accuracies using various training loss functions and distance metrics.
Main Methods:
- Leveraging hidden features extracted by BiGANs and CycleGANs through a kernel density estimation (KDE)-based inference method.
- Training the models to estimate the probability density function (PDF) of COVID-19 CT scans.
- Comparing classification performance against CAE-based feature extraction methods.
Main Results:
- The proposed CycleGAN-based models achieved approximately 16% higher classification accuracy than benchmark CAE-based models.
- BiGAN-based models showed an improvement of about 14% in classification accuracy compared to CAE-based models.
- Performance was evaluated across different training loss functions and distance metrics.
Conclusions:
- BiGANs and CycleGANs, when used for feature extraction, offer superior performance for COVID-19 detection in CT scans compared to CAE-based methods.
- The developed KDE-based inference engine effectively utilizes learned hidden features for disease detection.
- This study opens new avenues for applying generative models in medical diagnostic classification tasks.
